AI-300 Practice Question 81: Cluster Idle Time

Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 9 · question 1 of 10

Design and implement an MLOps infrastructureeasy

A compute cluster has a minimum of 0 nodes and an idle time before scale down of 120 seconds. The team submits training jobs about every 5 minutes during the day, and every job waits several minutes for a node to be provisioned. They still want the cluster to scale to zero overnight. What should they change?

Select one answer
Show answer and explanation

Answer:

  • D. Increase the idle time before scale down beyond the gap between jobs

After a job finishes, an idle node is kept for the configured idle time before the cluster scales down. If jobs arrive less than that interval apart, the node is reused and nobody waits for provisioning. With a minimum of 0 the cluster still scales down fully when the idle time passes overnight (option D).

- A is wrong because a maximum of 0 would prevent the cluster from running any job at all. - B is wrong because a smaller size does not stop nodes from being released between jobs, so provisioning delays would remain. - C is wrong because a compute instance is a single development VM and not a scalable target for queued jobs.

Reference: Create an Azure Machine Learning compute cluster

Practice Test 9 ›

Study guides for AI-300